MI

cloudyu Mixtral 34Bx2 MoE 60B iMat

MistralGGUF by NexesenexUpdated 2024-02-09
Download on Hugging Face
Parameters60.8B
Context195K
Layers60
Architecturellama
Smallest quant20.8 GB
Quick answer

cloudyu Mixtral 34Bx2 MoE 60B iMat needs about 27.2 GB of memory at Q3_K_XS (a 24.7 GB file) with an 8K context window, so it fits on a 32 GB GPU. This repository has 5 quants, from 20.8 GB (Q2_K_S) to 29.2 GB (Q3_K_M). Longer context needs more: at 32K tokens the KV cache alone adds about 8.1 GB.

cloudyu Mixtral 34Bx2 MoE 60B iMat is a 60.8B-parameter Mistral model in GGUF format, quantized by Nexesenex. Its trained context length is 195K tokens. The repository has 5 quant files, from 20.8 GB (Q2_K_S) to 29.2 GB (Q3_K_M).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K_S20.8 GB 23.3 GB
Too big Get file →
Q2_K22.4 GB 24.9 GB
Too big Get file →
Q3_K_XS24.7 GB 27.2 GB
Too big Get file →
Q3_K_S26.3 GB 28.8 GB
Too big Get file →
Q3_K_M29.2 GB 31.7 GB
Too big Get file →

What GPU do you need for cloudyu Mixtral 34Bx2 MoE 60B iMat?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.

  • 4 GB VRAM: no quant fits at an 8K context.
  • 6 GB VRAM: no quant fits at an 8K context.
  • 8 GB VRAM: no quant fits at an 8K context.
  • 12 GB VRAM: no quant fits at an 8K context.
  • 16 GB VRAM: no quant fits at an 8K context.
  • 24 GB VRAM: the best quant that fits is Q2_K_S (20.8 GB file, about 23.3 GB in use).
  • 32 GB VRAM: the best quant that fits is Q3_K_M (29.2 GB file, about 31.7 GB in use).

Which cloudyu Mixtral 34Bx2 MoE 60B iMat quant should you choose?

Q3_K_XS (24.7 GB) is the middle option in this repository. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.

If memory is tight, the smallest file is Q2_K_S at 20.8 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturellama
Parameters60.8B (approximate, mixture of experts)
Layers60
Trained context200,000 tokens
KV cache (fp16)240 KB per token

Source

Quantized byNexesenex
FamilyMistral
Downloads (HF)27
Last updated2024-02-09

Frequently asked questions about cloudyu Mixtral 34Bx2 MoE 60B iMat

How much VRAM does cloudyu Mixtral 34Bx2 MoE 60B iMat need?

About 27.2 GB at Q3_K_XS with an 8K context window. That is the 24.7 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.

Can I run cloudyu Mixtral 34Bx2 MoE 60B iMat on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 23.3 GB, so you would need a larger GPU or CPU offloading.

What is the file size of cloudyu Mixtral 34Bx2 MoE 60B iMat GGUF?

It depends on the quant: Q2_K_S is 20.8 GB, Q2_K is 22.4 GB, Q3_K_XS is 24.7 GB, Q3_K_S is 26.3 GB, Q3_K_M is 29.2 GB.

What context length does cloudyu Mixtral 34Bx2 MoE 60B iMat support?

The GGUF header lists a trained context length of 200,000 tokens (195K). The KV cache grows with the context you use: about 2.0 GB at 8K and 8.1 GB at 32K.

Where can I download cloudyu Mixtral 34Bx2 MoE 60B iMat GGUF?

The files are on Hugging Face in the Nexesenex/cloudyu_Mixtral_34Bx2_MoE_60B-iMat.GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

Similar models

Browse all →

More from Nexesenex

All by Nexesenex →

More Mistral models

All Mistral →

Other large models

All large models →